Revert to scikit-learn 0.19.1.

gama/gama.py

@@ -13,8 +13,7 @@
 import arff
 import pandas as pd
 import numpy as np
-from sklearn.preprocessing import OneHotEncoder
-from sklearn.impute import SimpleImputer
+from sklearn.preprocessing import OneHotEncoder, Imputer
 
 import gama.genetic_programming.compilers.scikitlearn
 from gama.genetic_programming.algorithms.metrics import Metric
@@ -345,7 +344,7 @@ def _preprocess_numpy(self, X, y):
         # This helps us use a wider variety of algorithms without constructing a grammar.
         # One should note that ideally imputation should not always be done since some methods work well without.
         # Secondly, the way imputation is done can also be dependent on the task. Median is generally not the best.
-        self._imputer = SimpleImputer(strategy="median")
+        self._imputer = Imputer(strategy="median")
         self._imputer.fit(X)
         if np.isnan(X).any():
             log.info("Feature matrix X has been found to contain NaN-labels. Data will be imputed using median.")
@@ -355,7 +354,7 @@ def _preprocess_numpy(self, X, y):
 
     def _construct_y_score(self, y):
         if any(metric.requires_probabilities for metric in self._metrics):
-            return OneHotEncoder(categories='auto').fit_transform(y.reshape(-1, 1)).todense()
+            return OneHotEncoder().fit_transform(y.reshape(-1, 1)).todense()
         return y
 
     def _search_phase(self, X, y, warm_start=False, restart_criteria=None, timeout=1e6):

gama/utilities/auto_ensemble.py

@@ -244,7 +244,7 @@ def __init__(self, metric, y_true, label_encoder=None, *args, **kwargs):
         self._label_encoder = label_encoder
 
         # For metrics that only require class labels, we still want to apply one-hot-encoding to average predictions.
-        self._one_hot_encoder = OneHotEncoder(categories='auto').fit(self._y_true.reshape(-1, 1))
+        self._one_hot_encoder = OneHotEncoder().fit(self._y_true.reshape(-1, 1))
 
         if self._metric.requires_probabilities:
             self._y_score = self._one_hot_encoder.transform(self._y_true.reshape(-1, 1)).toarray()

setup.py

@@ -5,7 +5,7 @@
 requirements = [
     'numpy>=1.14.0',
     'scipy>=1.0.0',
-    'scikit-learn==0.20.0',
+    'scikit-learn==0.19.1',
     'stopit>=1.1.1',
     'liac-arff>=2.2.2',
     'category-encoders>=1.2.8'